It's not a real problem. It's one of the most honest business models currently employed by tech companies - simple exchange of money for value. Training wasn't free and only few players on the planet had enough capital to perform it. Serving isn't free, it costs electricity (and maintenance). The companies that trained the models, and companies that serve inference, have all created real value at their own expense, and they're (for now) charging very little for it. The value users get from inference - that no one is even capturing right now, it's literally left on the table and goes 100% to the users.
Contrast that with most other businesses - tech or otherwise - where there's always strings and trickery attached to any transaction.
While I agree with you regarding shady business practices, you're very conveniently skipping over the fact that open source licenses _REQUIRE_ attribution.
> While I agree with you regarding shady business practices, you're very conveniently skipping over the fact that open source licenses _REQUIRE_ attribution.
When you use the code as is or create a derivative work. The knowledge embodied by the code and encapsulated in an LLM doesn't strike me as needing to give attribution because the code the LLM would product doesn't match any particular open source code base.
At least that's my thinking. I'd be curious to see an example where you think attribution is necessary and how you would actually do it given an output from an LLM.
I hear you, but I think you might miss my point, which is while LLMs are clearly trained on copyrighted material, what they produce (their output) is NOT a copy of a specific code snippet they were trained on in a way that you would say "that's a copy from this code base".
Thanks for that link to the definition and requirements for something to be considered a derivative work.
I think my interpretation, based on your link, holds: unless the LLM output (transformation) substantially bears the original source code author's creation and personality, there is nothing to give attribution to.
(Your argument reminded me of the oil industry and the (original argument for the) economic relationship between "Big Oil" and resource rich nations.)
> strings and trickery
One could argue that at this initial stage, just as with surveillance capitalism and services like "free email", the general public is being treated like the natives who sell their land for a few trinkets. Once we are passed this stage and just like other surveillance tech our social and economical life becomes effectively dependent on these services, we can review if we have not sold our future for some (arguably dubious) "value".
Go back to early '00s. How did you "value" the "transaction" of handing over the handling of your personal electronic correspondences to a corporate entity that may or may not be an extension of the security state?
I don't disagree. Fortunately, for as long as open-weight models continue to track SOTA with a few months lag, we lose at most those few months if big providers decide to stop playing nice with the people.
Aren't you paying attention to all the news? They have announced that this will be regulated. A technology that everyone is told "can kill us all" is going to be treated like WMDs. Enjoy those open source models while they are still legal.
So the material trained on has no value, but all other costs associate with ai should be captured by business? I mean we coders released it, and it’s hard to compensate everyone, but the world would be better if some open source projects got funded. They don’t even get a source footnote.
Does anyone trust Ai companies not to eventually spy on you and feed you ads? It’s not like we haven’t seen this playbook before.
I don't trust any company, AI or otherwise. Business is business, companies are only nice while margins are good.
I'm commenting on how things are now, not how they may turn out at some point in the future. This is in response to complains that are also mostly about now, not about hypothetical.
WRT LLMs in particular, open-weight models offset the risk a lot. As long as they track SOTA by couple months, that's the most we lose in capability should the commercial vendors start to enshittify their inference services.
Them complaining about distillation is 100% hypocritical, but also understandable; they found a goose laying golden eggs, but didn't expect it to be so easy to clone through distillation.
It's just ordinary business. You always have to get every benefit you can and deny everything you can to your competitors. The basic premise of free market economics is that when everyone does this, it will even out. OpenAI doing everything they can to stop DeepSeek, together with DeepSeek doing everything they can to stop OpenAI stopping them, is expected market behaviour. David Graeber wrote about this as his "goon" category of Bullshit Job.
Not tongue in cheek, but called that to distinguish it from ROM BASIC. AFAIK, Advanced BASIC had to be loaded from the DOS disk and mostly added disk management and graphics features.
I’m the same. I once suggested on the Transport Tycoon forums that there be a mechanism for curated sets of NewGRFs that work well together to be selected from the main page and was resolutely told that it’s up to the user to put together their own sets as they like.
EDIT: it looks like something like this is now in v16 as “collections”! I wonder if these can be shared.
I play the national lottery in the UK mainly because of the good causes it supports. Athletes competing in the Olympics for Team GB for example receive significant funding from the lottery. I see my ticket as a charity donation with the added fun of an astronomically small chance of winning money.
The iPhone did everything the Nokia 3310 did, better. Electric cars do not (yet) do better some things hybrids do, such as being able to be fuelled with 400+ miles of range in 5 minutes.
I’m nowhere near the point of wanting an electric car to replace my hybrid. The convenience of petrol and the cost of electricity is too high. High electricity costs aren’t going to be fixed in my country any time soon so Toyota will continue to have a huge market here.
Toyotas may need less repairs than other vehicles, but of course they have the same maintenance schedules and costs as other vehicles.
Compare periodic oil changes, spark plug changes, ignition coils, stolen catalytic converters, exhaust system, PCV system, air and fuel filters, brake pads, transmission fluid, and other ICE maintenance items with the electric drivetrain. At 120,000 km I've replaced the tires once and the brake pads look brand new. That's it. Even the windshield wipers are still in good shape for some reason.
Most EVs these days can recharge 300ish miles in 15 mins, but 99% of the time I don’t even have to drive anywhere to refuel as it get recharged overnight in my garage. EVs are waaay more efficient in terms of MPGe so at least for me it is less half in terms of cost to refuel compared to petrol not even considering the external cost of emitting CO2
15 minutes is still not as good as the 2 minutes it takes to fill up a gas tank. Electric cars just aren't there yet for long drives, though they are great for everyday driving around town.
The iPhone actually had way worse battery life than basically any Nokia. It’s a great comparison. People happily traded more features for having to plug their phone in every night.
iPhone also had same modem as everyone. It wasn't a Wi-Fi device, it was a phone-computer hybrid.
Compared to that, EVs feel more like Wi-Fi or WiMAX device that owners would say theirs are daily drivable but only make Discord calls. Overall situation more closely resemble PDAs before iPhone.
We’ve been driving them for over a decade now. They aren’t new anymore, and they still aren’t a panacea. There is this cool thing called car rental, it lets you use cars that aren’t your own.
I work for an online grocer, and I really do think it's not really an issue due to two things for us:
* the amount of stock going through one fulfillment center instead of landing on shelves in smaller stores, means we never have old products laying around. The cucumber you get from us came in a few hours ago. The one in your store has been laying there for days and touched by many. 10 stores each need their own buffer to handle variable demand and thus overstock and get deliveries for certain products rarely. We don't. Our spoilage is so so low compared to traditional stores.
* anyways, to alleviate the fears of ordering something that's about to expire, we guarantee x amount of days for perishable products.
Some supermarkets in the UK (e.g. Waitrose) literally just pick items already out and on display in the local store to give to delivery customers. So you’re getting whatever is at the front of the shelf in your local store, which is the least fresh.
What I found (but this was during the panny-D) that only grocery stuff had an even longer shelf life than the store itself, that is, it was really fast moving stuff apparently.
This is because space is _stiff_. Recall Hooke’s law from high school physics. The k constant represents the stiffness of the object. A rubber band is about 50. A sky scraper, about a million. Space? About 10^46 if I recall correctly. So it takes a truly enormous amount of energy in the form of gravitational waves to be able to move space enough for it to be detectable on Earth. And the only objects that can do that are the most massive ones moving at close to the speed of light: black holes, neutron stars, supernovae (the latter would have to be very close for us to see gravitational waves from - close enough that we’d likely see it with the naked eye as well).
reply